[ryux1]

Closes #16342.

Preserve the default integer result while propagating a custom operation's return type through rowcol and index.

Agent used: OpenAI Codex.

[thomas-maschler]

I don't think all the extra @overload functions are necessary here.

Use _T = TypeVar("_T", default=int)

and for the existing rowcol() use

op: _RoundOperation[_T] | None = None and return tuple[_T, _T] | tuple[list[_T], list[_T]]

You can remove all the extra overloads. This should address the issue.

[ryux1]

Updated to use a defaulted type variable and removed the extra overloads as suggested. I also applied the same simplification to TransformerBase.rowcol; the existing deprecated overload is retained only for the precision parameter. The full targeted stubs/rasterio test suite passes locally.

[ryux1]

Removed the added regression tests, kept the typing-only change, and rebased onto current main. The targeted rasterio stub suite passes across pre-commit, structure, Pyright, ty, pyrefly, and mypy.

[ryux1]

Apologies for the delayed follow-up. I have rebased the branch onto current main while preserving the requested simplification and removal of the regression-test files. The targeted rasterio suite now passes locally across pre-commit, repository structure, Pyright, ty, pyrefly, and mypy. The updated branch is ready for another look when convenient.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉

[thomas-maschler]

@JelleZijlstra can we merge this?

[srittau]

@thomas-maschler I think non-maintainers should also be able to review and approve PRs here. It definitely helps us if common contributors (who often have more experience than us for third-party libraries) do that. So please feel free to review and mark as "approved" if you think a PR is ready!

[thomas-maschler]

@srittau happy to help. I didn't have the option to approve this PR.

[srittau]

@thomas-maschler Too bad, I know it used to be possible, and the repository settings also allow it. Maybe that function has been disabled python org-wide. That said, a LGTM or similar will also do the job. (And your review above was certainly very helpful!)